Get Your Business Ready for Automation

Before you connect an automation, make it possible to tell which information it should trust. You don't need to organize every document in the business. Start with the records that one task actually uses.

Start with One Question the Task Must Answer

Suppose you want to sort incoming service inquiries. The task may need the requested service, location, contact details and whether the person is already a customer. It doesn't automatically need your entire customer history, financial records or staff files.

Write down the intended result first: "Put each inquiry into one service category and flag missing information for a person." Then list the information needed for that result. Leave unrelated records out.

Choose the Record That Wins

A price in an old PDF and a different price in a current spreadsheet create a decision the automation should not guess. For each important fact, name the record that should be used and the person responsible for keeping it current.

A small information register can have four columns: the fact, its approved source, its owner and its last check date. For example, "service areas / current service-area sheet / operations owner / checked this month." The date records a check; it doesn't prove the information will stay correct.

Archive or label old versions so they are not mistaken for current instructions. Don't delete business records just to make a folder tidy. Retention and accounting requirements need their own decision.

Make Missing Information Visible

Use the same field names and formats where you can. Keep a person's email separate from notes about their inquiry. Choose a consistent date format and include a time zone when the time matters. Give records stable identifiers so two people with the same name are not merged by guesswork.

Don't fill a blank field with a plausible answer. Decide what the task should do when a value is missing, conflicting or out of date. "Needs review" is a useful result when the alternative is an invented fact.

Prepare Examples Without Opening Everything

Create a small permitted test set with an ordinary inquiry, a missing field, a duplicate and a request outside your service area. Write the expected outcome beside each one.

Prefer invented examples when they test the same issue. If real records are necessary, check who may use them and remove information the test doesn't need. Removing a name alone may not remove identifying details from free-text notes.

A useful test asks whether the automation follows the selected source and stops at uncertainty. A clean-looking output isn't enough if it used the wrong record.

Leave with a Small, Checkable Pack

For this one task, you should now have an approved source list, field definitions, a rule for missing information and examples with expected results. Ask the person who owns the process to check it before connecting live actions.

Keep this first pack limited to the information the selected task needs.

Build Your First Information Register

Try the inquiry-sorting example with four rows: service requested, service area, existing-customer status and contact address. Beside each row, write the approved source and what happens if the value is unavailable. For service area, that might be "current service-area sheet; flag for review if the address isn't listed."

Now introduce a conflict: an old brochure says you serve a town that the current sheet excludes. The expected test result is a review flag or the outcome your authorized rule specifies. It shouldn't blend the two sources into a new policy.

Ask the process owner to work through your examples without explaining the missing details aloud. Anything you have to add verbally belongs in the register or instructions. NIST's AI risk framework provides wider context on assessing AI systems through their design, use and evaluation. Your register is a small working method, not a certification.

---